Vehicle Detection and Tracking Project
Extending Advances Lane Finding Project @ https://github.com/Matz89/CarND-Advanced-Lane-Lines
The goals / steps of this project are the following:
Perform a Histogram of Oriented Gradients (HOG) feature extraction on a labeled training set of images and train a classifier Linear SVM classifier
Optionally, you can also apply a color transform and append binned color features, as well as histograms of color, to your HOG feature vector.
Note: for those first two steps don't forget to normalize your features and randomize a selection for training and testing.
Implement a sliding-window technique and use your trained classifier to search for vehicles in images.
Run your pipeline on a video stream (start with the test_video.mp4 and later implement on full project_video.mp4) and create a heat map of recurring detections frame by frame to reject outliers and follow detected vehicles.
Estimate a bounding box for vehicles detected.
Import Packages
#CODE BLOCK 0
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np
import cv2
import glob
Read and display all Test Images
--using cv2.imread
Test Road Images
#CODE BLOCK 1
#image reader
test_img_path = ".\\test_images\\"
test_img_filename = '*'
test_images = glob.glob(test_img_path + test_img_filename)
#images contained here
img_shp = cv2.imread(test_images[0]).shape
test_imgs = np.empty((len(test_images), img_shp[0], img_shp[1], img_shp[2]), dtype="uint8")
#Display all images with matplotlib
figx = 2
figy = 3
f, axarr = plt.subplots(figx,figy, figsize=(20,10))
y = 0
for i, img in enumerate(test_images):
this_img = cv2.imread(img)
this_img_resized = cv2.resize(this_img,(img_shp[1], img_shp[0]))
test_imgs[i] = this_img_resized
x = i % figx
axarr[x,y].imshow(cv2.cvtColor(this_img, cv2.COLOR_BGR2RGB))
axarr[x,y].set_title(img.split('\\')[-1])
y = y+1 if x >= figx-1 else y
#Camera image details
img_w = img_shp[1]
img_h = img_shp[0]
img_ch = img_shp[2]
num_imgs = len(test_imgs)
print("Number of Images: {0}\nImage Width: {1}\nImage height: {2}\nColor Channels: {3}".format(num_imgs, img_w, img_h, img_ch))
#display images
#plt.tight_layout()
#plt.show()
Camera Calibration Images
#CODE BLOCK 2
#image reader
camera_img_path = ".\\camera_cal\\"
camera_img_filename = '*'
camera_images = glob.glob(camera_img_path + camera_img_filename)
#images contained here - use shape of first image - fixed sized for greater performance
img_shp = cv2.imread(camera_images[0]).shape
camera_imgs = np.empty((len(camera_images), img_shp[0], img_shp[1], img_shp[2]), dtype="uint8")
#Display all images with matplotlib
figx = 5
figy = 4
f, axarr = plt.subplots(figx,figy, figsize=(20,10))
y = 0
for i, img in enumerate(camera_images):
this_img = cv2.imread(img)
this_img_resized = cv2.resize(this_img,(img_shp[1], img_shp[0]))
camera_imgs[i] = np.uint8(this_img_resized)
x = i % figx
axarr[x,y].imshow(cv2.cvtColor(this_img, cv2.COLOR_BGR2RGB))
axarr[x,y].set_title(img.split('\\')[-1])
y = y+1 if x >= figx-1 else y
#Camera image details
img_w = img_shp[1]
img_h = img_shp[0]
img_ch = img_shp[2]
num_imgs = len(camera_imgs)
print("Number of Images: {0}\nImage Width: {1}\nImage height: {2}\nColor Channels: {3}".format(num_imgs, img_w, img_h, img_ch))
#display images
#plt.tight_layout()
#plt.show()
Helper Functions
Camera Calibration Functions
#CODE BLOCK 3
#For use with chessboard calibration only!
#return retval, undist, cornersImg
def calibrate_camera(calibration_imgs, nx, ny):
objp = np.zeros((nx*ny, 3), np.float32)
objp[:,:2] = np.mgrid[0:nx, 0:ny].T.reshape(-1, 2) #object x,y coordinates
objpoints = []
imgpoints = []
for i, src_img in enumerate(camera_imgs):
img = np.copy(src_img)
#We assume BGR format
#Convert to Grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
#Find chessboard corners
ret, corners = cv2.findChessboardCorners(gray, (nx, ny), None)
if ret == True:
#Draw and display the corners
cornersImg = cv2.drawChessboardCorners(img, (nx, ny), corners, ret)
objpoints.append(objp)
imgpoints.append(corners)
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], None, None)
return mtx, dist
#Example/Test
mtx, dist = calibrate_camera(camera_imgs, 9, 6)
for img in camera_imgs:
timg = np.copy(img)
undist = cv2.undistort(timg, mtx, dist, None, mtx)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
ax1.set_title('Original Image', fontsize=50)
ax2.imshow(cv2.cvtColor(undist, cv2.COLOR_BGR2RGB))
ax2.set_title("Undistorted Image", fontsize=50)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
Colour Transforms and Gradients
#CODE BLOCK 4
#Use Sobel and HLS thresholding to find edges in image
def findSobelxHLSThresh(src_img, s_thresh=(170,255), sx_thresh=(20,100), asColour=False):
img = np.copy(src_img)
#Covert to HLS color space and separate the channels
hls = cv2.cvtColor(img, cv2.COLOR_BGR2HLS)
l_ch = hls[:,:,1]
s_ch = hls[:,:,2]
#Sobel x
sobelx = cv2.Sobel(l_ch, cv2.CV_64F, 1, 0) #Take the derivative in x
abs_sobelx = np.absolute(sobelx) #Absolute x derivative to accentuate lines away from horizontal
scaled_sobel = np.uint8(255*abs_sobelx/np.max(abs_sobelx))
#Threshold x gradient
sxbinary = np.zeros_like(scaled_sobel)
sxbinary[(scaled_sobel >= sx_thresh[0]) & (scaled_sobel <= sx_thresh[1])] = 1
#Threshold color channel
s_binary = np.zeros_like(s_ch)
s_binary[(s_ch >= s_thresh[0]) & (s_ch <= s_thresh[1])] = 1
#Stack channels; Green = sobelx, Blue = Saturation
if asColour:
color_binary = np.dstack((np.zeros_like(sxbinary), sxbinary, s_binary)) * 255
else:
combined = np.add(sxbinary, s_binary)
color_binary = np.dstack((combined, combined, combined))
color_binary[color_binary > 0] = 255
return color_binary
#Example/Test
for i, img in enumerate(test_imgs):
edgeResult = findSobelxHLSThresh(img, asColour = True)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
ax1.set_title('Original Image', fontsize=40)
ax2.imshow(edgeResult)
ax2.set_title('Colour Result', fontsize=40)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
Perspective Transform
#CODE BLOCK 5
#Defined Region coordinates
TOP_LEFT_XY = [.44,.65]
TOP_RIGHT_XY = [.56,.65]
BOTTOM_LEFT_XY = [.20,.92]
BOTTOM_RIGHT_XY = [.80,.92]
#overlays image with polygon defined by vertices
def view_region_of_interest(img, vertices):
r_img = np.copy(img)
vertices = vertices[0]
prev = vertices[-1]
for v in vertices:
#creating a polygon defined by "vertices" with the fill color Blue
cv2.line(r_img,(prev[0], prev[1]),(v[0],v[1]),(255,0,0),5)
prev = v
return r_img
#Example/Test
tImage = test_imgs[0]
#Define Region of Interest (x,y)
v1 = np.multiply(np.flip(tImage.shape[:2], axis=0), BOTTOM_LEFT_XY) #Bottom Left
v2 = np.multiply(np.flip(tImage.shape[:2], axis=0), TOP_LEFT_XY) #Top Left
v3 = np.multiply(np.flip(tImage.shape[:2], axis=0), TOP_RIGHT_XY) #Top Right
v4 = np.multiply(np.flip(tImage.shape[:2], axis=0), BOTTOM_RIGHT_XY) #Bottom Right
verts = np.array([v1,v2,v3,v4], dtype=np.int32)
region_image = view_region_of_interest(tImage, [verts])
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(cv2.cvtColor(tImage, cv2.COLOR_BGR2RGB))
ax1.set_title('Original Image', fontsize=40)
ax2.imshow(cv2.cvtColor(region_image, cv2.COLOR_BGR2RGB))
ax2.set_title('Region Result', fontsize=40)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
#CODE BLOCK 6
#Offsets for perspective transform
w_offset = 300
h_offset = 50
def topDownPerspective(img, src_pts, dst_pts):
pImg = np.copy(img)
img_size = (pImg.shape[1], pImg.shape[0])
M = cv2.getPerspectiveTransform(src_pts, dst_pts)
warped = cv2.warpPerspective(pImg, M, img_size, flags=cv2.INTER_LINEAR)
return warped, M
#Example/Test
for i, img in enumerate(test_imgs):
shpy = img.shape[0]
shpx = img.shape[1]
img_size = (shpx, shpy)
src_pts = np.float32(verts)
dst_tl = [w_offset,h_offset]
dst_tr = [shpx-w_offset,h_offset]
dst_bl = [w_offset,shpy - h_offset]
dst_br = [shpx-w_offset,shpy-h_offset]
dst_pts = np.float32([dst_bl, dst_tl, dst_tr, dst_br])
undist_img = cv2.undistort(img, mtx, dist, None, mtx)
edge_img = findSobelxHLSThresh(undist_img, asColour=False)
result_img, result_M = topDownPerspective(edge_img,src_pts, dst_pts )
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
ax1.set_title('Original Image', fontsize=40)
ax2.imshow(result_img, cmap='gray')
ax2.set_title('TopDown Result', fontsize=40)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
Lane Finding
#CODE BLOCK 7
def identifyLanes(src_img, nwindows=9, margin=100, minpix=50):
#nwindowsl Choose the number of sliding windows
#margin; Set the width of the windows +/- margin
#minpix; Set the minimum number of pixels found to recenter window
#Prep copy of image
img = np.copy(src_img)
img = np.sum(img, axis=-1)
img[img > 0] = 255
#collection of rectangle tuples
rectangles = []
# Take a histogram of the bottom half of the image
histogram = np.sum(img[img.shape[0]//2:,:], axis=0)
# Create an output image to draw on and visualize the result
out_img = np.dstack((img,img,img))
out_img = out_img.astype("uint8")
# Find the peak of the left and right halves of the histogram
# These will be the starting point for the left and right lines
midpoint = np.int(histogram.shape[0]//2)
leftx_base = np.argmax(histogram[:midpoint])
rightx_base = np.argmax(histogram[midpoint:]) + midpoint
# Set height of windows
window_height = np.int(img.shape[0]//nwindows)
# Identify the x and y positions of all nonzero pixels in the image
nonzero = img.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Current positions to be updated for each window
leftx_current = leftx_base
rightx_current = rightx_base
# Create empty lists to receive left and right lane pixel indices
left_lane_inds = []
right_lane_inds = []
# Step through the windows one by one
for window in range(nwindows):
# Identify window boundaries in x and y (and right and left)
win_y_low = img.shape[0] - (window+1)*window_height
win_y_high = img.shape[0] - window*window_height
win_xleft_low = leftx_current - margin
win_xleft_high = leftx_current + margin
win_xright_low = rightx_current - margin
win_xright_high = rightx_current + margin
#Pack away rectangle points to show areas of interest - can be drawn later
rectangles.append(((win_xleft_low,win_y_low),(win_xleft_high,win_y_high)))
rectangles.append(((win_xright_low,win_y_low),(win_xright_high,win_y_high)))
# Identify the nonzero pixels in x and y within the window
good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xleft_low) & (nonzerox < win_xleft_high)).nonzero()[0]
good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xright_low) & (nonzerox < win_xright_high)).nonzero()[0]
# Append these indices to the lists
left_lane_inds.append(good_left_inds)
right_lane_inds.append(good_right_inds)
# If you found > minpix pixels, recenter next window on their mean position
if len(good_left_inds) > minpix:
leftx_current = np.int(np.mean(nonzerox[good_left_inds]))
if len(good_right_inds) > minpix:
rightx_current = np.int(np.mean(nonzerox[good_right_inds]))
# Concatenate the arrays of indices
left_lane_inds = np.concatenate(left_lane_inds)
right_lane_inds = np.concatenate(right_lane_inds)
# Extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
#Colouring lanes left=Red, right=Blue
out_img[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
out_img[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
#Generating x and y values for plotting
ploty = np.linspace(0, out_img.shape[0]-1, out_img.shape[0] )
left_fitx = left_fit[0]*ploty**2 + left_fit[1]*ploty + left_fit[2]
right_fitx = right_fit[0]*ploty**2 + right_fit[1]*ploty + right_fit[2]
return out_img, left_fit, right_fit, rectangles, (left_fitx, right_fitx, ploty)
#Example/Test
for i, img in enumerate(test_imgs):
shpy = img.shape[0]
shpx = img.shape[1]
img_size = (shpx, shpy)
src_pts = np.float32(verts)
dst_tl = [w_offset,h_offset]
dst_tr = [shpx-w_offset,h_offset]
dst_bl = [w_offset,shpy - h_offset]
dst_br = [shpx-w_offset,shpy-h_offset]
dst_pts = np.float32([dst_bl, dst_tl, dst_tr, dst_br])
undist_img = cv2.undistort(img, mtx, dist, None, mtx)
edge_img = findSobelxHLSThresh(undist_img, asColour=False)
topdown_img, topdown_M = topDownPerspective(edge_img,src_pts, dst_pts )
result_img, left_fit, right_fit, rects, img_plots = identifyLanes(topdown_img)
ploty = img_plots[2]
left_fitx = img_plots[0]
right_fitx = img_plots[1]
#Draw Rectangles for output image
for pts in rects:
cv2.rectangle(result_img,pts[0],pts[1],(0,255,0), 2)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(topdown_img, cmap='gray')
ax1.set_title('TopDown View', fontsize=40)
ax2.imshow(result_img)
ax2.set_title('Identified Lanes', fontsize=40)
ax2.plot(left_fitx, ploty, color='yellow')
ax2.plot(right_fitx, ploty, color='yellow')
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
#CODE BLOCK 8
def identifyLanesNext(src_img, left_fit, right_fit, margin=100):
#If providing left_fit/right_fit polynomials (ie after finding the lanes in the previous frame with identifyLanes)
#This will find new points within the margin of the left_fit/right_fit lines
out_img = np.copy(src_img)
nonzero = out_img.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
left_lane_inds = ((nonzerox > (left_fit[0]*(nonzeroy**2) + left_fit[1]*nonzeroy +
left_fit[2] - margin)) & (nonzerox < (left_fit[0]*(nonzeroy**2) +
left_fit[1]*nonzeroy + left_fit[2] + margin)))
right_lane_inds = ((nonzerox > (right_fit[0]*(nonzeroy**2) + right_fit[1]*nonzeroy +
right_fit[2] - margin)) & (nonzerox < (right_fit[0]*(nonzeroy**2) +
right_fit[1]*nonzeroy + right_fit[2] + margin)))
# Again, extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
# Generate x and y values for plotting
ploty = np.linspace(0, out_img.shape[0]-1, out_img.shape[0] )
left_fitx = left_fit[0]*ploty**2 + left_fit[1]*ploty + left_fit[2]
right_fitx = right_fit[0]*ploty**2 + right_fit[1]*ploty + right_fit[2]
#Colouring lanes left=Red, right=Blue
out_img[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
out_img[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
return out_img, left_fit, right_fit, (left_fitx, right_fitx, ploty)
Measuring Curvature
#CODE BLOCK 9
def findCurvatureOfLane(ploty, leftx, rightx):
# Define conversions in x and y from pixels space to meters
ym_per_pix = 30/720 # meters per pixel in y dimension
xm_per_pix = 3.7/700 # meters per pixel in x dimension
# Define y-value where we want radius of curvature
# I'll choose the maximum y-value, corresponding to the bottom of the image
y_eval = np.max(ploty)
# Fit new polynomials to x,y in world space
left_fit_cr = np.polyfit(ploty*ym_per_pix, leftx*xm_per_pix, 2)
right_fit_cr = np.polyfit(ploty*ym_per_pix, rightx*xm_per_pix, 2)
# Calculate the new radii of curvature
left_curverad = ((1 + (2*left_fit_cr[0]*y_eval*ym_per_pix + left_fit_cr[1])**2)**1.5) / np.absolute(2*left_fit_cr[0])
right_curverad = ((1 + (2*right_fit_cr[0]*y_eval*ym_per_pix + right_fit_cr[1])**2)**1.5) / np.absolute(2*right_fit_cr[0])
return left_curverad, right_curverad
#CODE BLOCK 10
def findDevFromCenter(ploty, leftfit, rightfit, car_pos_x, img_h):
# Define conversions in x and y from pixels space to meters
xm_per_pix = 3.7/700 # meters per pixel in x dimension
left_x_val = leftfit[0]*img_h**2 + leftfit[1]*img_h + leftfit[2]
right_x_val = rightfit[0]*img_h**2 + rightfit[1]*img_h + rightfit[2]
#Lane Center position
lane_center_x = (left_x_val + right_x_val) / 2
car_deviation = (car_pos_x - lane_center_x) * xm_per_pix
return car_deviation
Vehicle Detection and Tracking
Reading in Vehicle and Non-Vehicle Images for Training/Testing
#CODE BLOCK 11
#image reader
training_img_path = ".\\training\\"
non_vehicle_img_filenames = 'non-vehicles\\*\\*.png'
vehicle_img_filenames = 'vehicles\\*\\*.png'
non_vehicle_images = glob.glob(training_img_path + non_vehicle_img_filenames)
vehicle_images = glob.glob(training_img_path + vehicle_img_filenames)
#images contained here
img_shp = cv2.imread(non_vehicle_images[0]).shape
non_vehicle_imgs = np.empty((len(non_vehicle_images), img_shp[0], img_shp[1], img_shp[2]), dtype="uint8")
#Display random images with matplotlib
figx = 2
figy = 4
f, axarr = plt.subplots(figx,figy, figsize=(20,10))
y = 0
for i, veh_img in enumerate(vehicle_images):
this_veh_img = cv2.imread(veh_img)
this_non_veh_img = cv2.imread(non_vehicle_images[i])
x = i % figx
axarr[0,y].imshow(cv2.cvtColor(this_veh_img, cv2.COLOR_BGR2RGB))
axarr[0,y].set_title("Vehicle Image")
axarr[1,y].imshow(cv2.cvtColor(this_non_veh_img, cv2.COLOR_BGR2RGB))
axarr[1,y].set_title("Non-Vehicle Image")
y = y+1 if x >= figx-1 else y
if y>=figy:
break
#Training image details
num_veh_imgs = len(vehicle_images)
num_nonveh_imgs = len(non_vehicle_images)
print("Number of Vehicle Images: {0}\nNumber of Non-Vehicle Images: {1}".format(num_veh_imgs, num_nonveh_imgs))
#display images
plt.tight_layout()
plt.show()
Extracting Histogram of Oriented Gradients (HOG)
#CODE BLOCK 12
from skimage.feature import hog
def get_hog_features(img, orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True):
# Call with two outputs if vis==True
if vis == True:
features, hog_image = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
block_norm= 'L2-Hys',
cells_per_block=(cell_per_block, cell_per_block),
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features, hog_image
# Otherwise call with one output
else:
features = hog(img, orientations=orient,
pixels_per_cell=(pix_per_cell, pix_per_cell),
cells_per_block=(cell_per_block, cell_per_block),
block_norm= 'L2-Hys',
transform_sqrt=True,
visualise=vis, feature_vector=feature_vec)
return features
#Example/Test
#Let random index for vehicle image/Non-Vehicle image
v_ind = np.random.randint(0, num_veh_imgs)
nv_ind = np.random.randint(0, num_nonveh_imgs)
#Read Images
v_img = cv2.imread(vehicle_images[v_ind])
nv_img = cv2.imread(non_vehicle_images[nv_ind])
#Get Grayscale of Images
v_gray = cv2.cvtColor(v_img, cv2.COLOR_BGR2GRAY)
nv_gray = cv2.cvtColor(nv_img, cv2.COLOR_BGR2GRAY)
# Call our function with vis=True to see an image output
v_features, v_hog_image = get_hog_features(v_gray, orient= 9,
pix_per_cell= 8, cell_per_block= 2,
vis=True, feature_vec=False)
nv_features, nv_hog_image = get_hog_features(nv_gray, orient= 9,
pix_per_cell= 8, cell_per_block= 2,
vis=True, feature_vec=False)
f, axarr = plt.subplots(2,2, figsize=(20,10))
axarr[0,0].imshow(v_gray, cmap='gray')
axarr[0,0].set_title("Vehicle Image")
axarr[0,1].imshow(nv_gray, cmap='gray')
axarr[0,1].set_title("Non-Vehicle Image")
axarr[1,0].imshow(v_hog_image, cmap='gray')
axarr[1,0].set_title("HOG Visualization (Vehicle)")
axarr[1,1].imshow(nv_hog_image, cmap='gray')
axarr[1,1].set_title("HOG Visualization (Non-Vehicle)")
Extracting Colour Histogram Features
#CODE BLOCK 13
def color_hist(img, nbins=32, bins_range=(0, 256)):
#Separate colour channels and find histograms
channel1_hist = np.histogram(img[:,:,0], bins=nbins, range=bins_range)
channel2_hist = np.histogram(img[:,:,1], bins=nbins, range=bins_range)
channel3_hist = np.histogram(img[:,:,2], bins=nbins, range=bins_range)
#Concat into single vector
hist_features = np.concatenate((channel1_hist[0], channel2_hist[0], channel3_hist[0]))
return hist_features
#Example/Test
#Let random index for vehicle image/Non-Vehicle image
v_ind = np.random.randint(0, num_veh_imgs)
nv_ind = np.random.randint(0, num_nonveh_imgs)
#Read Images
v_img = cv2.imread(vehicle_images[v_ind])
nv_img = cv2.imread(non_vehicle_images[nv_ind])
#Colour Histograms
v_hists = color_hist(v_img)
nv_hists = color_hist(nv_img)
v_rh = v_hists[:32]
v_gh = v_hists[32:64]
v_bh = v_hists[64:]
nv_rh = nv_hists[:32]
nv_gh = nv_hists[32:64]
nv_bh = nv_hists[64:]
bin_edges = np.histogram(v_img[:,:,0], bins=32, range=(0,256))[1]
bin_centers = (bin_edges[1:] + bin_edges[0:len(bin_edges)-1])/2
fig = plt.figure(figsize=(12,3))
plt.subplot(141)
plt.imshow(cv2.cvtColor(v_img, cv2.COLOR_BGR2RGB))
plt.title("IMAGE")
plt.subplot(142)
plt.bar(bin_centers, v_rh)
plt.xlim(0, 256)
plt.title('R Histogram')
plt.subplot(143)
plt.bar(bin_centers, v_gh)
plt.xlim(0, 256)
plt.title('G Histogram')
plt.subplot(144)
plt.bar(bin_centers, v_bh)
plt.xlim(0, 256)
plt.title('B Histogram')
fig.tight_layout()
fig = plt.figure(figsize=(12,3))
plt.subplot(141)
plt.imshow(cv2.cvtColor(nv_img, cv2.COLOR_BGR2RGB))
plt.title("IMAGE")
plt.subplot(142)
plt.bar(bin_centers, nv_rh)
plt.xlim(0, 256)
plt.title('R Histogram')
plt.subplot(143)
plt.bar(bin_centers, nv_gh)
plt.xlim(0, 256)
plt.title('G Histogram')
plt.subplot(144)
plt.bar(bin_centers, nv_bh)
plt.xlim(0, 256)
plt.title('B Histogram')
fig.tight_layout()
Extracting Spatial Bin Features
#CODE BLOCK 14
def bin_spatial(img, size=(32, 32)):
#Ravel all colour channels then stack
color1 = cv2.resize(img[:,:,0], size).ravel()
color2 = cv2.resize(img[:,:,1], size).ravel()
color3 = cv2.resize(img[:,:,2], size).ravel()
return np.hstack((color1, color2, color3))
#Examples/Test
#Let random index for vehicle image/Non-Vehicle image
v_ind = np.random.randint(0, num_veh_imgs)
nv_ind = np.random.randint(0, num_nonveh_imgs)
#Read Images
v_img = cv2.imread(vehicle_images[v_ind])
nv_img = cv2.imread(non_vehicle_images[nv_ind])
#extract spatial feature
v_spatial = bin_spatial(v_img)
nv_spatial = bin_spatial(nv_img)
#plot for visual
fig = plt.figure(figsize=(20,4))
plt.subplot(121)
plt.plot(v_spatial)
plt.title("Vehicle Spatial Features")
plt.subplot(122)
plt.plot(nv_spatial)
plt.title("Non-Vehicle Spatial Features")
fig.tight_layout()
Extract features of a multiple images
#CODE BLOCK 15
from sklearn.svm import LinearSVC
from sklearn.preprocessing import StandardScaler
def convert_colour(img, conv='BGR'):
#Colour Conversion
if color_space != 'BGR':
if color_space == 'HSV':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
elif color_space == 'LUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2LUV)
elif color_space == 'HLS':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2HLS)
elif color_space == 'YUV':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2YUV)
elif color_space == 'YCrCb':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2YCrCb)
elif color_space == 'RGB':
feature_image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
else: feature_image = np.copy(img)
return feature_image
def extract_features(img_list, colour_space='BGR', spatial_size=(32, 32),
hist_bins=32, orient=9,
pix_per_cell=8, cell_per_block=2, hog_channel=0,
spatial_feat=True, hist_feat=True, hog_feat=True):
#Contain all features
features = []
for file in img_list:
#Init empty list to contain image features
img_features = []
img = cv2.imread(file)
feature_image = convert_colour(img, colour_space)
#Spatial Features
if spatial_feat == True:
spatial_features = bin_spatial(feature_image, size=spatial_size)
img_features.append(spatial_features)
#Histogram Features
if hist_feat == True:
hist_features = color_hist(feature_image, nbins=hist_bins)
img_features.append(hist_features)
#HOG Features
if hog_feat == True:
if hog_channel == 'ALL':
hog_features = []
for channel in range(feature_image.shape[2]):
hog_features.extend(get_hog_features(feature_image[:,:,channel],
orient, pix_per_cell, cell_per_block,
vis=False, feature_vec=True))
else:
hog_features = get_hog_features(feature_image[:,:,hog_channel], orient,
pix_per_cell, cell_per_block, vis=False, feature_vec=True)
img_features.append(hog_features)
features.append(np.concatenate(img_features))
return features
Train the model
#CODE BLOCK 16
#TUNE-ABLE PARAMETERS
color_space = 'YCrCb' # Can be RGB, HSV, LUV, HLS, YUV, YCrCb
orient = 9 # HOG orientations
pix_per_cell = 8 # HOG pixels per cell
cell_per_block = 2 # HOG cells per block
hog_channel = 'ALL' # Can be 0, 1, 2, or "ALL"
spatial_size = (16, 16) # Spatial binning dimensions
hist_bins = 32 # Number of histogram bins
spatial_feat = True # Spatial features on or off
hist_feat = True # Histogram features on or off
hog_feat = True # HOG features on or off
y_start_stop = [400, 650] # Min and max in y to search in slide_window()
#CODE BLOCK 17
import time
from sklearn.svm import LinearSVC
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
print('Extracting features...')
t=time.time()
vehicle_features = extract_features(vehicle_images, colour_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
non_vehicle_features = extract_features(non_vehicle_images, colour_space=color_space,
spatial_size=spatial_size, hist_bins=hist_bins,
orient=orient, pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
hog_channel=hog_channel, spatial_feat=spatial_feat,
hist_feat=hist_feat, hog_feat=hog_feat)
t2 = time.time()
print(round(t2-t, 2), 'Seconds to extract features...')
#Stack features
print(len(vehicle_features))
print(len(non_vehicle_features))
X = np.vstack((vehicle_features, non_vehicle_features)).astype(np.float64)
#Init labels and stack same order as above
y = np.hstack((np.ones(len(vehicle_features)), np.zeros(len(non_vehicle_features))))
#Split into training and test sets (random)
rand_state = np.random.randint(0, 100)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=rand_state)
# Fit a per-column scaler
X_scaler = StandardScaler().fit(X_train)
# Apply the scaler to X
X_train = X_scaler.transform(X_train)
X_test = X_scaler.transform(X_test)
print('Using:',orient,'orientations',pix_per_cell,'pixels per cell and', cell_per_block,'cells per block')
print('Feature vector length:', len(X_train[0]))
print('Number of Features:', len(X_train))
# Use a linear SVC
svc = LinearSVC()
# Check the training time for the SVC
t=time.time()
svc.fit(X_train, y_train)
t2 = time.time()
print(round(t2-t, 2), 'Seconds to train SVC...')
# Check the score of the SVC
print('Test Accuracy of SVC = ', round(svc.score(X_test, y_test), 4))
# Check the prediction time for a single sample
t=time.time()
Save trained data
#CODE BLOCK 18
from sklearn.externals import joblib
model_pickle = {}
model_pickle["svc"] = svc
model_pickle["scaler"] = X_scaler
model_pickle["orient"] = orient
model_pickle["pix_per_cell"] = pix_per_cell
model_pickle["cell_per_block"] = cell_per_block
model_pickle["spatial_size"] = spatial_size
model_pickle["hist_bins"] = hist_bins
model_pickle["hist_feat"] = hist_bins
model_pickle["spatial_feat"] = spatial_feat
model_pickle["hog_feat"] = hist_bins
model_pickle["hog_channel"] = hog_channel
model_pickle["color_space"] = color_space
joblib.dump(model_pickle, 'model.pkl')
Load trained data
#CODE BLOCK 19
from sklearn.externals import joblib
model = joblib.load('model.pkl')
Heatmap Functions
#CODE BLOCK 20
from scipy.ndimage.measurements import label
def add_heat(heatmap, bbox_list):
# Iterate through list of bboxes
for box in bbox_list:
# Add += 1 for all pixels inside each bbox
# Assuming each "box" takes the form ((x1, y1), (x2, y2))
heatmap[box[0][1]:box[1][1], box[0][0]:box[1][0]] += 1
# Return updated heatmap
return heatmap# Iterate through list of bboxes
def apply_threshold(heatmap, threshold):
# Zero out pixels below the threshold
heatmap[heatmap <= threshold] = 0
# Return thresholded map
return heatmap
def draw_labeled_bboxes(img, labels):
# Iterate through all detected cars
for car_number in range(1, labels[1]+1):
# Find pixels with each car_number label value
nonzero = (labels[0] == car_number).nonzero()
# Identify x and y values of those pixels
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Define a bounding box based on min/max x and y
bbox = ((np.min(nonzerox), np.min(nonzeroy)), (np.max(nonzerox), np.max(nonzeroy)))
# Draw the box on the image
cv2.rectangle(img, bbox[0], bbox[1], (0,0,255), 6)
# Return the image
return img
def decay_heat(heatmap, rate):
#Decay all values in heatmap by rate;
heatmap -= rate
return heatmap
Find vehicles in image
#CODE BLOCK 21
# Define a single function that can extract features using hog sub-sampling and make predictions
def find_vehicles(img, ystart, ystop, scale, svc, X_scaler, orient, pix_per_cell, cell_per_block, spatial_size,
hist_bins, color_space='BGR', hist_feat=True, spatial_feat=True, hog_feat = True,
hog_channel = 'ALL'):
draw_img = np.copy(img)
img_features = []
bbox_list = []
#img_heatmap = np.zeros_like(draw_img[:,:,0]).astype(np.float)
img_tosearch = img[ystart:ystop,:,:]
#Colour Conversion
ctrans_tosearch = convert_colour(img_tosearch, conv=color_space)
if scale != 1:
imshape = ctrans_tosearch.shape
ctrans_tosearch = cv2.resize(ctrans_tosearch, (np.int(imshape[1]/scale), np.int(imshape[0]/scale)))
ch1 = ctrans_tosearch[:,:,0]
ch2 = ctrans_tosearch[:,:,1]
ch3 = ctrans_tosearch[:,:,2]
# Define blocks and steps as above
nxblocks = (ch1.shape[1] // pix_per_cell) - cell_per_block + 1
nyblocks = (ch1.shape[0] // pix_per_cell) - cell_per_block + 1
nfeat_per_block = orient*cell_per_block**2
# 64 was the orginal sampling rate, with 8 cells and 8 pix per cell
window = 64
nblocks_per_window = (window // pix_per_cell) - cell_per_block + 1
cells_per_step = 2 # Instead of overlap, define how many cells to step
nxsteps = (nxblocks - nblocks_per_window) // cells_per_step + 1
nysteps = (nyblocks - nblocks_per_window) // cells_per_step + 1
if hog_feat:
# Compute individual channel HOG features for the entire image
hog1 = get_hog_features(ch1, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog2 = get_hog_features(ch2, orient, pix_per_cell, cell_per_block, feature_vec=False)
hog3 = get_hog_features(ch3, orient, pix_per_cell, cell_per_block, feature_vec=False)
for xb in range(nxsteps):
for yb in range(nysteps):
ypos = yb*cells_per_step
xpos = xb*cells_per_step
if hog_feat:
hog_feat1 = hog1[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat2 = hog2[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_feat3 = hog3[ypos:ypos+nblocks_per_window, xpos:xpos+nblocks_per_window].ravel()
hog_features = np.hstack((hog_feat1, hog_feat2, hog_feat3))
xleft = xpos*pix_per_cell
ytop = ypos*pix_per_cell
# Extract the image patch
subimg = cv2.resize(ctrans_tosearch[ytop:ytop+window, xleft:xleft+window], (64,64))
# Get color features
if spatial_feat:
spatial_features = bin_spatial(subimg, size=spatial_size)
if hist_feat:
hist_features = color_hist(subimg, nbins=hist_bins)
#append all features in appropriate order as model
img_features.extend(spatial_features)
img_features.extend(hist_features)
img_features.extend(hog_features)
# Scale features and make a prediction
test_features = X_scaler.transform(np.hstack((spatial_features, hist_features, hog_features)).reshape(1, -1))
test_prediction = svc.predict(test_features)
if test_prediction == 1:
xbox_left = np.int(xleft*scale)
ytop_draw = np.int(ytop*scale)
win_draw = np.int(window*scale)
top_left = (xbox_left, ytop_draw+ystart)
if(top_left[0] > 700):
bottom_right = (xbox_left+win_draw,ytop_draw+win_draw+ystart)
cv2.rectangle(draw_img, top_left, bottom_right,(0,0,255),6)
#Used for heatmapping in output
bbox_list.append((top_left, bottom_right))
return draw_img, bbox_list
Example with Test Images
#CODE BLOCK 22
#Extracting saved parameters of model
svc = model["svc"]
X_scaler = model["scaler"]
orient = model["orient"]
pix_per_cell = model["pix_per_cell"]
cell_per_block = model["cell_per_block"]
spatial_size = model["spatial_size"]
hist_bins = model["hist_bins"]
hist_feat = model["hist_feat"]
spatial_feat = model["spatial_feat"]
hog_feat = model["hog_feat"]
hog_channel = 'ALL'
color_space = model["color_space"]
print(type(svc))
print(type(X_scaler))
print("Pix per cell ", pix_per_cell)
print("cell per block ", cell_per_block)
print("spatial_size ", spatial_size)
print("hist_bins ", hist_bins)
print("hist_feat ", hist_feat)
print("spatial_feat ", spatial_feat)
print("hog_feat ", hog_feat)
print("hog channel ", hog_channel)
print("color_space ", color_space)
heatmap_threshold = 3
scale = 1.0
hold_frames = 30
#CODE BLOCK 23
# Draw bounding boxes
def draw_boxes(img, bboxes, color=(0, 0, 255), thick=6):
imcopy = np.copy(img)
# Iterate through the bounding boxes
for bbox in bboxes:
cv2.rectangle(imcopy, bbox[0], bbox[1], color, thick)
# Return the image copy with boxes drawn
return imcopy
f, axarr = plt.subplots(len(test_imgs),3, figsize=(60,60))
for i, srcimg in enumerate(test_imgs):
img = np.copy(srcimg)
window_img, bbox_list = find_vehicles(img, y_start_stop[0], y_start_stop[1], scale, svc, X_scaler,
orient, pix_per_cell, cell_per_block,spatial_size, hist_bins,
color_space=color_space, hist_feat=hist_feat, spatial_feat=spatial_feat,
hog_feat=hog_feat, hog_channel='ALL')
heatmap = np.zeros_like(img[:,:,0]).astype(np.float)
heatmap = add_heat(heatmap, bbox_list)
heatmap = apply_threshold(heatmap, heatmap_threshold)
heatmap = np.clip(heatmap, 0, 255)
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(img), labels)
axarr[i][0].imshow(cv2.cvtColor(draw_img, cv2.COLOR_BGR2RGB))
axarr[i][1].imshow(cv2.cvtColor(window_img, cv2.COLOR_BGR2RGB))
axarr[i][2].imshow(heatmap)
plt.tight_layout()
plt.show()
Main Pipeline to Find Lanes
#CODE BLOCK 24
#REDEFINE Region Coordinates
TOP_LEFT_XY = [.44,.65]
TOP_RIGHT_XY = [.56,.65]
BOTTOM_LEFT_XY = [.20,.92]
BOTTOM_RIGHT_XY = [.80,.92]
#Offsets for perspective transform
w_offset = 300
h_offset = 50
#Camera Calibrations
mtx, dist = calibrate_camera(camera_imgs, 9, 6)
#Main pipeline to find lanes and return augmented image displaying lane information
class showLane:
def __init__(self):
self.left_fit = None
self.right_fit = None
self.heatmap = None
self.smoothing_counter = 0
self.boxes = []
self.heatmaps = []
def __call__(self, src_img):
#do not edit original image
img = np.copy(src_img)
line_width=50
#Image shape information
shpy = img.shape[0]
shpx = img.shape[1]
img_size = (shpx, shpy)
#Source points for perspective Transform (And region of interest for Lanes)
v1 = np.multiply(np.flip(img.shape[:2], axis=0), BOTTOM_LEFT_XY) #Bottom Left
v2 = np.multiply(np.flip(img.shape[:2], axis=0), TOP_LEFT_XY) #Top Left
v3 = np.multiply(np.flip(img.shape[:2], axis=0), TOP_RIGHT_XY) #Top Right
v4 = np.multiply(np.flip(img.shape[:2], axis=0), BOTTOM_RIGHT_XY) #Bottom Right
src_pts = np.float32(verts)
#Destination points for perspective transform
dst_tl = [w_offset,h_offset]
dst_tr = [shpx-w_offset,h_offset]
dst_bl = [w_offset,shpy - h_offset]
dst_br = [shpx-w_offset,shpy-h_offset]
dst_pts = np.float32([dst_bl, dst_tl, dst_tr, dst_br])
#Undistort image
undist_img = cv2.undistort(img, mtx, dist, None, mtx)
#Push image through threshhold filtering
thresh_img = findSobelxHLSThresh(undist_img, asColour=False)
#Create Top-Down perspective of relevant area of image
topdown_img, topdown_M = topDownPerspective(thresh_img,src_pts, dst_pts )
if(self.left_fit != None) and (self.right_fit != None):
fit_img, left_fit, right_fit, img_plots = identifyLanesNext(topdown_img, self.left_fit, self.right_fit)
self.left_fit = left_fit
self.right_fit = right_fit
else:
fit_img, left_fit, right_fit, rects, img_plots = identifyLanes(topdown_img)
self.left_fit = left_fit
self.right_fit = right_fit
#separate out img_plots
ploty = img_plots[2]
left_fitx = img_plots[0]
right_fitx = img_plots[1]
#Find curvature of lanes; radius in meters
left_curverad, right_curverad = findCurvatureOfLane(ploty, left_fitx, right_fitx)
#Find vehicle deviation from center of lane
car_dev = findDevFromCenter(ploty, self.left_fit, self.right_fit, img.shape[1]/2, img.shape[0])
#Create overlay image
out_img_faded = np.zeros_like(fit_img)
out_img = np.zeros_like(fit_img)
#Create Lane Line Visual Info
#True line width
line_width = int(line_width / 2)
#recast the x and y points into usable format for cv2.fillPoly()
left_line_window1 = np.array([np.transpose(np.vstack([left_fitx-line_width, ploty]))])
left_line_window2 = np.array([np.flipud(np.transpose(np.vstack([left_fitx+line_width, ploty])))])
left_line_pts = np.hstack((left_line_window1, left_line_window2))
right_line_window1 = np.array([np.transpose(np.vstack([right_fitx-line_width, ploty]))])
right_line_window2 = np.array([np.flipud(np.transpose(np.vstack([right_fitx+line_width, ploty])))])
right_line_pts = np.hstack((right_line_window1, right_line_window2))
#Create Lane Region Visual Info
#region_pts = np.concatenate((left_line_pts, np.flipud(right_line_pts)))
region_left = np.array(np.transpose(np.vstack([left_fitx, ploty])))
region_right = np.array(np.transpose(np.vstack([right_fitx, ploty])))
region_pts = np.concatenate((region_left, np.flipud(region_right)))
#Create Lane Curvature Visual Info (Average of both lanes)
curve = np.array(np.transpose(np.vstack([(left_fitx+right_fitx)/2, ploty]))).astype("int")
#Find a point along the line (near the bottom of the image
curve_point = curve[-1]
text_offset = 25;
text_pos = (curve_point[0]+text_offset, curve_point[1]-text_offset)
mid_curve_r = (left_curverad + right_curverad)/2
#Draw Visuals
region_colour = (0,255,0)
left_lane_colour = (255,0,0)
right_lane_colour = (0,0,255)
curve_colour = (155,255,255)
text_colour = (55,55,55)
cv2.fillPoly(out_img_faded, [region_pts.astype('int32')], region_colour)
cv2.fillPoly(out_img_faded, np.int_([left_line_pts]), left_lane_colour)
cv2.fillPoly(out_img_faded, np.int_([right_line_pts]), right_lane_colour)
start_pt = curve[0]
for end_pt in curve[1:]:
cv2.line(out_img, (start_pt[0], start_pt[1]), (end_pt[0], end_pt[1]), text_colour, 5)
start_pt = end_pt
#Return perspective to original view
persp_img_faded, out_M = topDownPerspective(out_img_faded,dst_pts, src_pts )
persp_img, out_M = topDownPerspective(out_img,dst_pts, src_pts )
#Overlay with original image
persp_img = cv2.addWeighted(persp_img_faded, 0.5, persp_img, 1.0, 0.0)
out_img = cv2.addWeighted(persp_img, 1.0, undist_img, 1.0, 0.0)
#Drawn interface information
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(out_img, "%.2fm (r.)" % mid_curve_r, text_pos, font, 1.0, curve_colour, 5)
cv2.putText(out_img, "dev from cntr: %.2fm" % car_dev, (15, 45), font, 1.0, curve_colour, 5)
window_img, bbox_list = find_vehicles(img, y_start_stop[0], y_start_stop[1], scale, svc, X_scaler,
orient, pix_per_cell, cell_per_block,spatial_size, hist_bins,
color_space=color_space, hist_feat=hist_feat, spatial_feat=spatial_feat,
hog_feat=hog_feat, hog_channel=hog_channel)
heat = np.zeros_like(img[:, :, 0]).astype(np.float)
heatmap = add_heat(heat, bbox_list)
self.heatmaps.append(heatmap)
if(len(self.heatmaps) > hold_frames):
self.heatmaps = self.heatmaps[-hold_frames:]
decay_rate = 1.0
else:
decay_rate = 0.75
average_heat = sum(self.heatmaps)//len(self.heatmaps)
average_heat = average_heat * decay_rate
out_heatmap = apply_threshold(average_heat, heatmap_threshold)
labels = label(out_heatmap)
out_img = draw_labeled_bboxes(np.copy(out_img), labels)
return out_img
#Example/Test
#Image Test
t_img = test_imgs[4]
tt = showLane()
r_img = tt(t_img)
#Plot Example Image
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(cv2.cvtColor(t_img, cv2.COLOR_BGR2RGB))
ax1.set_title('TopDown View', fontsize=40)
ax2.imshow(cv2.cvtColor(r_img, cv2.COLOR_BGR2RGB))
ax2.set_title('Pipeline Test', fontsize=40)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
Run pipeline with easy video sample
#CODE BLOCK 25
# Import everything needed to edit/save/watch video clips
from moviepy.editor import VideoFileClip
from IPython.display import HTML
#CODE BLOCK 26
inp_vid = "./project_video.mp4"
out_vid = "./output_videos/out_project_video.mp4"
clip1 = VideoFileClip(inp_vid)
out_clip = clip1.fl_image(showLane())
out_clip.write_videofile(out_vid, audio=False)